Stock Market Volatility Spillover from Developed Markets to Regional Markets Tiffany Grosvenor and Kevin Greenidge 1

Transcription

1 Sock Marke Volailiy Spillover from Developed Markes o Regional Markes Tiffany Grosvenor and Kevin Greenidge 1 ABSTRACT This paper builds on he work of Kim and Langrin (1996) o invesigae he co-movemen in sock markes beween he developing counries of he Caribbean as well as from developed markes. Mulivariae Generalised Auoregressive Condiional Heeroscedasicy (GARCH) is employed o examine he volailiy spillover beween he hree regional exchanges namely ha of Jamaica, Trinidad and Barbados, and from he New York Sock Exchange (NYSE). The sudy uilises daily daa on he composie index of each sock marke o assess he exen o which volailiy spillovers exis. Resuls sugges ha significan spillovers indeed exis beween each of he regional exchanges, as well as from he NYSE. Keywords: Caribbean, Sock Exchange, GARCH, Volailiy Spillover 1 Corresponding auhor: Kevin Greenidge, Research and Economic Analysis Deparmen, Cenral Bank of Barbados, Tom Adams Financial Cenre, Bridgeown, Barbados. Tel: ; Fax: ; kcgreenidge@cenralbank.org.bb

2 1. Inroducion The las wo o hree decades have winessed increasing levels of financial inegraion among he World economies, as resricions on capial mobiliy across counries have gradually weakened. The consequenial increase in cross-border financial flows, along wih he increasing regionalisaion of economic aciviy, has resuled in greaer inerdependence of major financial markes all over he world. Caribbean counries are no excepion here, being par of a regional grouping, many of hese counries have insiued a coninuing policy of financial marke liberalisaion, which should have resuled in increasing inernaionalisaion of heir financial markes. In addiion, hese counries have srong economic ies wih USA hrough inernaional rade. As such, one would expec some degree of informaion spillover beween he individual markes and also wih ha of he USA, which could be in erms of acual reurns, he volailiy of reurns or boh 1. Mos of he research has been focused on he more developed markes of he US and Europe, wih a focus on examining he exen of financial inegraion (See Eun and Shim 1989, Hamao e al. 1990, Lucey and Voronkova 2006, Chelley-Seeley 2005, among ohers). However, here has also been some recen work on emerging markes, for insance, Harrison and Moore (2009). The auhors invesigae he comovemen of sock markes beween he emerging economies of Cenral and Easern Europe and he developed Markes of Wesern Europe using ime-varying realised correlaion raios, ime-varying coinegraion saisics, and a Generalized Auoregressive Condiional Heroscedasiciy (GARCH) model. The resuls sugges an absence of significan comovemen beween he sock exchanges and show ha he developed equiy markes are more inegraed han emerging markes. While here has been less aenion in lieraure regarding he comovemen of sock markes in he Caribbean, a few sudies have invesigaed oher areas of volailiy concerning hese regional markes. For insance, Hamilon (1996) invesigaes he GARCH effec in he reurns of hree 1 Much of he earlier research concenraed exclusively on spillover of he firs momen, ha is, co-movemen among he reurns. However, more recen research have demonsraed ha ha much of he informaion would be revealed in he volailiy of sock prices, which is in he condiional second momens of he price, raher han in he price iself.

3 companies lised on he Jamaica Sock Exchange (JSE). The sudy esed he hypohesis found by Lamorex and Lasrapes (1990) ha he unobservable cause of volailiy is he ime varying raes of informaional arrival, and ha once volume is inroduced he GARCH effec is significanly reduced, implying ha volume migh be a good proxy of he informaional arrival. The auhor concludes ha GARCH does no always model reurn volailiy on he JSE well, and ha volume raded does no necessarily manifes he ime varying volailiy in sock reurns. In esimaing he responsiveness of secoral sub index reurns o changes in he domesic marke porfolio on he Trinidad and Tobago Sock Exchange, Leon e al. (2000) compares predicions of non-sysemaic risk, using GARCH and EGARCH specificaions of he error variance. The resuls sugges ha reurns for he porfolios of Commercial Banks and Conglomeraes respond more han proporionaely o changes in he marke porfolio, and ha nonsysemaic volailiy appears o have been greaer during periods of macroeconomic insabiliy and poliical unres. Of more relevance for our purposes is he sudy by Hurdi (2004). The sudy applies he GARCH-BEKK procedure o he reurns from he Jamaican bond, foreign exchange and sock markes in order o esimae he magniude of he common marke and cross-marke volailiy ransmission. The resuls of he model indicae ha here are generally high levels of common marke volailiy ransmission relaive o he cross marke volailiy ransmission. The auhors noe ha he srong common marke ransmission in he foreign marke and he sock marke relaive o he bond marke reflecs he uncerainy momenum ha ofen characerizes hese risky markes. The sronges cross marke effecs occur from he bond marke o he foreign exchange and sock markes. Kim and Langrin (1996) noe ha as conrols on capial movemens, including repariaion of he invesmen proceeds, is relaxed, i becomes easier for foreign and domesic invesors o move asses ino and ou of hese small emerging markes. The auhors use GARCH models o examine he quesion of wheher here is increased volailiy spillover from developed markes o he sock markes of Trinidad and Tobago and Jamaica as a resul of he liberalisaion of heir foreign exchange markes. The resuls sugges ha volailiy spillovers increased following he liberalizaion of he exchange marke in Jamaica, bu no for Trinidad. The reason for his was

4 argued o be ha he barriers o enry o he sock marke in Jamaica were more binding ha in Trinidad. In his sudy, we build on he work of Kim and Langrin (1996) o examine he exen volailiy spillover from a US sock exchange o ha of he regional markes, using he mos recen daa. In addiion, we analyse he degree of comovemen beween he regional markes, an area which was no invesigaed in he aforemenioned sudy. We also inroduce he sock marke of Barbados in he analysis, since o he auhors knowledge here are no previous sudies on he volailiy of he sock make in Barbados. Moreover, since hen, here has been some expansion and improvemen as i relaes o mulivariae GARCH models. Examining he comovemen beween sock markes is considered o be imporan for several reasons. Inernaional porfolio diversificaion is beneficial only if reurns from inernaional sock markes are no significanly correlaed (Harrison and Moore 2009). Bekaer (1995) found ha he emerging marke reurns are higher, and more predicable, wih higher volailiy ha developed markes and correlaions wih developed markes were low, hus represening aracive hedging opporuniies for invesors in developed markes. Sock marke comovemen also gives a measure of he level of marke inegraion beween he counries (Kim and Langrin 1996). Policy makers are also ineresed in wheher sock markes exhibi comovemen because in a world of increasingly liberalised capial flows, he degree of sock marke comovemen can impac on he sabiliy of he inernaional moneary sysem (Harrison and Moore 2009). Finally, analysing price volailiy can give marke paricipans an assessmen of he risk associaed wih various financial producs and hus faciliae heir valuaion along wih he developmen of differen hedging echniques (Ng, 2000). Hence, he fundamenal aim of he paper is o conduc an up-o-dae review on he exen of comovemen of volailiy beween sock markes wihin he Caribbean, as well as beween hese regional markes and more developed markes. The sudy employs daily reurns of each of he sock marke indices and uilises boh univariae and mulivariae GARCH models. The remainder of he paper is srucured as follows: secion 2 provides a brief background on he

5 regional sock markes; secion 3 describes he daa and oulines he mehodology; secion 4 presens and analyses he resuls; and secion 5 summarises and concludes. 2. Background on Regional Sock Markes The sock exchanges of Barbados, Trinidad and Tobago and Jamaica are he only sock markes wihin he Caribbean Region. In 1991, he hree exchanges enered ino an agreemen for cross border rading in equiy, o form he Regional Sock Exchange. The Regional Sock Exchange is no a physical eniy bu an agreemen of cooperaion o faciliae he purchase and sale of shares cross border. 2.1 Barbados The Barbados Sock Exchange (BSE), formerly he Securiies Exchange of Barbados, was esablished in 1987, under he Securiies Exchange Ac, Cap 318A, of 1982, in order o creae a marke o promoe rading in financial securiies and encourage invesmen by he public in business enerprises. The BSE was reincorporaed in 2001 simulaneously wih he enacmen of he Securiies Ac , which repealed and replaced he original Ac of I is a privaely owned (by is Members), non-profi organizaion. Also in 2001, he BSE swiched from he manual, open aucion oucry mehod of rading o elecronic rading using he Order rouing mehod. The Regular marke is he main marke of he BSE, while he Junior marke caers o smaller and newer public companies, which may no mee he requiremens necessary for lising on he Regular marke. The BSE is he smalles of he regional exchanges and here are approximaely 24 companies and 26 securiies currenly lised on he exchange, wih a marke capializaion for he composie index close o 5.5 billion (US dollars). The main indices on he BSE are he Local Share Index, he Cross Lised Index and he Composie Index.

6 2.2 Jamaica The Jamaica Sock Exchange (JSE) was incorporaed as a privae limied company in Augus Sock-rading on he JSE is resriced o Broker-members who rade boh as agens and as principals. For he firs ime in Jamaica s hisory, a US dollar share was lised on he JSE in July The Jamaica Cenral Securiies Deposiory was esablished in 1998, and since hen he back office operaions has been auomaed. Since January 2000, he Jamaican Sock Exchange had a fully auomaed rading plaform. 2.3 Trinidad and Tobago The Securiies Marke Trinidad and Tobago exised informally for abou weny (20) years before he official opening of he Trinidad and Tobago Sock Exchange. In he early 1970 s, he Governmen decided as a maer of policy o localise he foreign-owned commercial banking and manufacuring secors of he economy, a policy which was o allow such companies o dives and sell a majoriy of heir shares o naionals. The esablishmen of he sock exchange in 1981 under he provisions on he Securiies Indusry Ac 1981 was a naural exension of he policy o formalise he Securiies marke in Trinidad and Tobago. I was subsequenly replaced wih he Securiies Indusry Ac of 1995, o deal wih he inefficiencies of he former, which brough ino operaion he esablishmen of a Securiies and Exchange Commission. The Trinidad and Tobago Sock Exchange (TTSE) implemened he Horizon Elecronic Trading Sysem on March 18 h 2005 replacing he manual open oucry sysem which was used since is incepion. As a April 1 s 2008, he rading days for he Exchange are every business day, Monday o Friday and here are presenly hiry-nine (39) companies lised on he exchange, wih 41 securiies being raded. The composie index regisers curren a marke capializaion of approximaely 12 billion (US dollars).

7 3. Daa and Mehodology 3.1 Daa The sudy employs daily daa on composie indices on he Barbados Sock Exchange (BSE), he Jamaica Sock Exchange (JSE), he Trinidad Sock Exchange (TTSE) and he New York Sock Exchange (NYSE), from he period 2005 o This daa was obained from he respecive online daabases of he Sock Exchanges. Table 1 provides summary saisics of he daily reurns for he sample period. Daily reurns are calculaed as, ln, /, 100, where, is he index of he i h counry, in year on rading day d. Figures 1 and 2 plo he index for each sock marke and he daily reurns respecively. Table 1: Summary Saisics of Daily Reurns of he Regional Exchanges Mean Median Max Min Sd.Dev. Skew Kur Jarque-Bera p-value TTSE JSE BSE NYSE Mean reurns were negaive hroughou he period for he TTSE, JSE and BSE, he lowes being in Trinidad ( percen), while he average for he more developed NYSE was recorded a percen. In addiion o having he larges daily reurns, he NYSE is also significanly more volaile han he oher markes as indicaed by he higher sandard deviaion. Of he regional markes, average volailiy is higher on he JSE (0.807) as compared o for he BSE and for he TTSE. Wih he excepion of he sock marke in Jamaica, he oher markes are negaively skewed, and significanly so for he BSE. Addiionally, he kurosis of he reurns from each exchange are considerably higher han 3 suggesing a non-normal disribuion of reurns. This observaion is confirmed by he significance of he Jarque-Bera saisic in each case.

8 Figure 1 Composie Indices of he Regional Exchnages ,400 1,200 1, /1/2005 4/1/2006 4/1/2007 4/1/2008 4/1/2009 4/1/2010 NYSE TTSE JSE BSE Figure 2 Daily Reurns o he Regional Exchanges TTSE JSE BSE NYSE

9 Table 2 presens he conemporaneous and lagged cross-correlaion coefficiens of reurns beween he exchanges and indicaes ha he cross correlaions beween each of he reurns are relaively low. A similar resul was also found by Kim and Langrin (1996) who examined he cross correlaions in boh pre- and pos-liberalisaion periods of he TTSE, he JSE and he Sandard and Poors 500 (S&P) index. However, he coefficiens appear o be somewha higher for some pairs of reurns han ha found by he auhors, even in he pos-liberalisaion period. For insance, cross correlaion coefficens beween he JSE and NYSE ranging beween and were found in Kim and Langrin (1996), compared o o In addiion he highes correlaions were found beween he JSE and he TTSE, a resul which was also observed in he above-menioned sudy. Table 2: Cross-correlaions Beween JSE, TTSE, BSE and NYSE Reurns Lags on JSE TTSE BSE NYSE NYSE JSE TTSE Anoher obsevaion is ha he cross correlaions beween he regional exchanges remains a he same level up o a lag lengh of wo, whereas correlaion wih he NYSE is more shor-erm as i becomes smaller afer he firs lag. 3.2 Mehodology Alhough here are several approaches o assessing he exen of he spillover of informaion among he naional markes, ARCH family of models is prehaps he mos popular as i perains o financial ime series. As such, his paper uilises mulivariae GARCH o examine he

10 volailiy spill over effecs amongs he regional sock markes, as well as he spill over from he more developed NYSE o hese markes o deermine he exen of co-movemen. ARCH models were inroduced by Engle (1982), and generalized as GARCH (Generalised ARCH) by Bollerslev (1986) and Taylor (1986). The GARCH model saes ha he error variance of any series y a ime depends on he squared error erms from previous periods as well as pas variances, where he ARCH effec (he coefficien on he pas squared error erms) capures he shor-run persisence of innovaions (an indicaion of he srengh of he shocks in he shor run) and he GARCH effec (he coefficien on he pas variances) measures he conribuion of hese innovaions o long run persisence Univariae Condiional Volailiy Models Consider he simple GARCH (1, 1) specificaion for daily reurns: Y = ' θ + ε (1) X ε = η 2 2 σ = ω + αε σ, η ~ iid(0,1) βσ 2 1 (2) where equaion (1) is he mean equaion given as a funcion of exogenous variables and an error erm, and ω >0, α >0 and β >0 are sufficien condiions o ensure ha he condiional variance 2 σ >0. The parameer α represens he ARCH effec, while β represens he GARCH effec. Maximum Likelihood esimaion (MLE) is used o obain he parameers of he model wih a join normal disribuion ofη. Weiss (1986) and Bollerslev and Wooldridge (1992) show ha he quasi-maximum likelihood esimaor (QMLE) is consisen if he condiional mean and condiional variance are correcly specified. However, he QMLE is inefficien, wih he degrees of inefficiency increasing wih he degree of deparure from normaliy (Engle and Gonzalez- Rivera 1991). As a resul, oher disribuions, namely he Suden and GED disribuions, are

11 available which would capure he higher observed kurosis, and hus perform beer. The Skewed Suden disribuion could also be used o accoun for skewness, since he abovemenioned disribuions assume symmery. The necessary and sufficien condiion for he exisence of saionariy for he GARCH (1,1) model isα + β <1, which is also sufficien for he consisency of QMLE. Nelson (1991) derived he weaker log-momen condiion which is also sufficien for QMLE given as: 2 E [log( αη + β)] < 0 (3) However, he QMLE is inefficien, wih he degrees of inefficiency increasing wih he degree of deparure from normaliy (Engle and Gonzalez-Rivera 1991). The observed excess kurosis in each of he series sugges faer ails han for a normal disribuion Equaion (2) assumes ha he effecs of posiive shocks on he condiional variance of daily reurns are he same as negaive shocks, i.e. a symmeric GARCH model. In order o capure possible asymmery in he daa, Glosen e al. (1993) pu forward he asymmeric (or hreshold) GARCH, or GJR model, given by: where = ω + ( α + γi( η 1)) ε 1 + βσ 1 σ (4) 1, η < 0 I( η ) = (5) 0, η 0 is an indicaor variable which differeniaes beween posiive and negaive shocks such ha asymmeric effecs are capured by γ, and ω >0, α + λ >0 and β >0, are sufficien condiions o ensure he posiiviy of he condiional variance. The shor run persisence of a posiive shock o daily sock reurns is given byα, whereas he persisence of a negaive shock is given by ( α + γ ). However, given ha he condiional shocks follow a symmeric disribuion, he average shor run persisence of shocks is α + γ / 2, while he conribuion of shocks o average long run persisence is α + λ / 2 + β. The necessary and sufficien condiion for he exisence of a second momen of he GJR (1,1) model is given as α + β + γ / 2 <1. McAleer (2003) gives he weaker log momen condiion for he GJR (1,1) as:

12 2 E [(log(( α + γi ( η )) η + β )] < 0 (6) This sudy explores he GARCH (1,1), AR(1)-GARCH (1,1), and ARMA(1,1)-GARCH(1,1) univariae specificaions o deermine which models fi he daa bes. The ARMA parameers are included in he mean equaions o accoun for possible serial correlaion in sock reurns, which could be as a resul of infrequen or non-synchronous rading. The respecive GJR models are also included o capure likely asymmery in he daa as i is expeced ha a negaive shock o sock reurns would resul in greaer volailiy. To deermine he appropriaeness of applying he GARCH mehodology, he ARCH es is used o es for he presence of condiional heeroscedasiciy in each of he mean equaions. The resuls sugges ha ARCH effecs indeed exis in he mean equaions of each of he four (4) series Mulivariae Condiional Volailiy Models This secion uilises he mulivariae GARCH (MGARCH) process o assess he exen of covolailiy among he daily reurns of he BSE, TTSE and JSE, and co-volailiy beween reurns on he NYSE and each of he regional exchanges. Consider he following specificaion for daily reurns: y = E( y I 1 ) + ε ε ~ N(0, H ) (7) where y is a mx1 vecor of daily reurns for each exchange, and Ι is an mxm marix of hisorical informaion available a ime. ε is a mx 1 vecor, which is independenly and idenically disribued (iid ) wih mean 0 and variance be specified as H. Similar o he univariae model, he variance may H 2 = + α * ε 1 + β * H 1 ω (8) whereω represens a mxm marix of consans, α is a mxm marix of coefficiens measuring ARCH effecs, while β is a mxm marix capuring GARCH effecs. Hence, he volailiy of, and co-volailiy among, he varying sock markes can be modeled using equaion (8).

13 The mos common approaches o mulivariae GARCH are he VECH model of Bollerslev, Engle, and Wooldridge (1988) and he BEKK model of Baba e. al (1990). The problem wih he general MGARCH models oulined above is ha he size of he variance-covariance marix increases exponenially as we increase he number of variables in he model, making he general model exremely difficul o esimae. H To reduce he compuaional burden of GARCH modeling, one can impose a diagonal VECH (DVEC) model or a diagonal BEKK (DBEKK) model, or alernaively use a Consan Condiional Correlaion (CCC) model (Bollerslev 1990). The CCC model imposes a consan correlaion over ime, while he diagonal BEKK model does no capure spillover. Hence, given he objecive of his paper, he DVEC approach is uilised. The DVEC ransformaion requires ha innovaions, ε and lagged volailiy H 1. H depends on he squares and cross producs of H = C + A* 1 ε 1 + B * H 1 ε (9) VEC is he operaor ha sacks he lower riangle of he variance-covariance marix. A is a mxm marix of ARCH erms and B is a mxm marix of GARCH erms measuring own volailiy and cross volailiy spill over. As wih he univariae specificaion, he parameers are subjeced o he posiiviy condiions imposed on he MGARCH process and co-variance saionariy is required. In his paper, equaion 9 is esimaed in he Eviews programme version 7.0 using boh he BHHH and Marquand algorihms, depending on which is able o achieve convergence. 4. Analysis of Resuls 4.1 Univariae Condiional Volailiy Models The resuls of he univariae GARCH and GJR models for daily reurns of he sock indices are presened in Tables 3 and 4 respecively. The condiional mean esimaes sugges ha on average he ARMA (1,1)-GARCH (1,1) esimaes are preferred for each of he sock markes, on he

14 basis of he significance of esimaes, he Akaike Informaion Crierion and he Schwarz Informaion Crierion. Addiionally, he Log-likelihood is improved in he ARMA (1,1) specificaion relaive o he ohers. The resuls of he misspecificaion ess, including he Box- Pierce saisics, sugges ha generally he models are well specified. The sufficien condiions, ω >0, α >0 and β >0 o ensure posiiviy are me for each of he series, as well as he second momen condiions. Hence, he coefficiens of he condiional variance for hese models are consisen and asympoically normal and inference on hese esimaes can be used for policy analysis. The coefficiens of he variance equaions are saisically significan boh in he shorand he long-run for each sock marke, indicaing he presence of imporan ARCH effecs. Based on he resuls obained from he ARMA (1, 1) specificaion, he ARCH coefficiens range beween and These resuls indicae ha he degree of own volailiy spillover is highes for he JSE and lowes for BSE. Volailiy persisence (he GARCH effec), which can be inerpreed as he markes susain he pas volailiy changes of sock reurns o he volailiy in he fuure periods, is highes for he NYSE (0.913) compared o 0.855, and for he TTSE, BSE and JSE respecively. Table 3 repors he asymmeric GARCH models, namely he GJR (1,1), AR (1)-GJR (1,1) and ARMA(1,1)-GJR(1,1) models. The coefficiens of he mean equaions bare similar saisical significance o he GARCH models. The necessary condiion has been me o ensure posiiviy, as well as he second momen condiion. The asymmeric effec is significan, for each of he sock reurns (and posiive) suggesing ha he effec of negaive shocks on condiional volailiy is greaer han posiive shocks. This resul is no surprising since i is expeced ha an anicipaed fall in prices on he sock marke would creae greaer uncerainy, han an unanicipaed increase. Therefore, a pas negaive shock in he mean sock reurns o each of he respecive markes would have a greaer impac on oday s volailiy, when compared o a pas posiive shock. This effec is larges for he NYSE, relaive o he oher markes, which may be aribued o more efficien ransmission of informaion, and which would allow companies o respond o news more quickly.

15 Table 3: Univariae GARCH (1,1), AR(1)-GARCH(1,1) and ARMA(1,1)-GARCH(1,1) Esimaes Counry Mean Equaion Variance Equaion C AR(1) MA(1) ωˆ αˆ βˆ AIC SIC Log Likelihood BSE 0.023* 0.027** 0.055** 0.749** * ** 0.055** 0.748** ** ** 0.027** 0.054** 0.747** JSE ** 0.062** 0.928** ** 0.063** 0.928** ** ** 0.087** 0.166** 0.726** TTSE ** 0.008** 0.085** 0.851** * 0.142** 0.007** 0.078** 0.859** ** ** 0.008** 0.070** 0.855** NYSE 0.051* 0.014** 0.081** 0.911** ** ** 0.013** 0.079** 0.914** ** 0.709** ** 0.013** 0.080** 0.913** Noe: ** and * indicaes significance a he 5 and 10 percen levels of esing. Counry Table 4: Univariae GJR (1,1), AR(1)-GJR(1,1) and ARMA(1,1)-GJR(1,1) Esimaes Mean Equaion Variance Equaion C AR(1) MA(1) ωˆ αˆ γ βˆ AIC SIC Log Likelihood BSE ** ** 0.139** 0.770** ** 0.026** 0.044** 0.760** ** ** 0.025** 0.029** 0.037** 0.760** JSE ** 0.032** 0.051** 0.930** ** 0.032** 0.052** 0.929** ** ** 0.032** 0.055** 0.930** TTSE * 0.008** 0.072** 0.022** 0.851** * 0.147** 0.007** 0.063** 0.027** 0.859** ** ** 0.008** 0.050** 0.034** 0.858** NYSE ** ** 0.144** 0.941** ** * 0.135** 0.941** ** ** 0.131** 0.939** Noe: ** and * indicaes significance a he 5 and 10 percen levels of esing.

16 4.2 Mulivariae Condiional Volailiy Models Given ha he ARMA(1,1) specificaion appears o fi he daa bes, we uilise he VARMA (1,1)-GARCH(1,1) approach o capure co-movemen in he volailiy of mean reurns o each of he sock markes. The esimaes for he condiional mean equaion of he model are provided in Table 5. Table 5: VARMA(1,1)-GARCH(1,1) Condiional Mean Esimaes Coefficien Sandard Error Coefficien Sandard Error BSE (i=1) JSE (i=2) ω ** ** α i ** α i ** α i ** ** α i ** TTSE (i=3) NYSE (i=4) ω ** α i α i * α i ** α i ** ** Noe: ** and * indicaes significance a he 5 and 10 percen levels of esing. The resuls sugges ha here are mean spillover effecs beween he regional markes, as well as from he NYSE. In Barbados, lagged reurns on he TTSE influence curren reurns, while he JSE had an insignifican impac on reurns in his counry. Lagged reurns on he TTSE have a significan effec on he JSE, and vice versa. However, evidence suggess negaive mean spillover from he TTSE o he JSE, while he JSE appears o have a posiive impac on he TTSE. The NYSE has a mean posiive spillover on he hree regional exchanges, however, he inra-regional mean effecs are larger han hose originaing from he NYSE, perhaps because of he direc link beween he regional exchanges due o he number of cross lisings. The esimaed condiional variance-covariance equaions are presened in Table 6. The resuls sugges ha here is evidence of significan own-volailiy spillover for each series of reurns as previously indicaed by he univariae resuls. Wih regards o he cross volailiy spillover, here

17 is evidence of ransmission beween each pair of regional markes, as well as beween he regional markes and ha of he NYSE. Parameer Table 6: VARMA(1,1)-GARCH(1,1) Condiional Variance Esimaes BSE (i=1) JSE (i=2) TTSE (i=3) NYSE(i=4) Coefficien Coefficien Coefficien Coefficien ω i ** ω i ** ω i ** ω i ** α i ** α i ** 0.104** α i ** 0.067** 0.043** α i ** 0.081** 0.052** 0.063** β i ** β i ** 0.792** β i ** 0.831** 0.872** β i ** 0.859** 0.901** 0.931** α ii +β ii Noe: ** and * indicaes significance a he 5 and 10 percen levels of esing. As i relaes o persisence in his condiional volailiy ransmission, he GARCH effecs are also significan beween he four sock exchanges. Own-volailiy persisence bears similar resuls o ha suggesed by he univariae specificaion. The coefficiens for cross-volailiy persisence range beween (beween he BSE and JSE) and 0.901(beween he NYSE and TTSE). The highes level of inra-regional cross volailiy persisence occurs beween he JSE and TTSE (0.831), while he highes level of cross volailiy persisence from he NYSE is eviden in he TTSE. 5. Conclusion The purpose of his paper is o examine he volailiy spillover beween he sock markes of he Caribbean, namely he JSE, he TTSE and he BSE, and o deermine he exen of spillover from he developed NYSE o hese regional markes. Using univariae and mulivariae GARCH echniques, he sudy employs daily daa from 2005 o 2008 on he composie sock marke indices o evaluae he exen of co-movemen of volailiy of reurns.

18 The univariae resuls sugges ha here are indeed ARCH and GARCH effecs in each of he series, and ha he ARMA specificaion appears o be superior. Own-volailiy spillover was found o be highes for he JSE and own-volailiy persisence being highes in he NYSE. Asymmeric effecs were found o be significan in each of he series indicaing ha a negaive shock o he volailiy of reurns oday would have a greaer impac on volailiy omorrow han a posiive shock. The findings also sugges ha here are significan mean spillover effecs among he hree Caribbean counries and also beween he NYSE and he exchanges in hese counries. In paricular, he evidence suggess significan bi-direcional mean spillover beween he TTSE and he JSE, while he BSE had no significan impac on eiher of he wo regional exchanges. Lagged reurns on he NYSE influences curren reurns on each of he regional markes, bu o a much lesser degree han he inra-regional spillover. As far as volailiy ransmission is concerned, here are significan cross spillover effecs beween each of he series. The GARCH effecs sugges ha he highes level of inra-regional cross volailiy persisence occurs beween he JSE and he TTSE, while he highes level eviden from he NYSE is eviden in he TTSE. The BSE has he lowes level of ransmission from he NYSE when compared o he oher regional markes, which is no surprising since i is he smalles of he hree markes. Addiionally, Trinidad and Tobago and Jamaica have engaged in more foreign exchange marke liberalisaion procedures, relaive o Barbados, which would sugges a greaer level of financial inegraion. These findings sugges ha he regional markes are sufficienly inegraed so ha each marke responds o boh curren and hisorical news generaed in he oher markes of he Caribbean, an o a lesser exen developed markes. Collecively he findings imply ha invesmen and fund managers wih access o news on oher regional markes may reac o changes faser han hose who do no. In addiion, he resuls also imply ha invesors should no only rely curren regional news o guide heir invesmen decisions bu also ake ino consideraion hisorical regional and inernaional news.

19 References Agbeyegbe, T. D Some Sylised Facs abou he Jamaica Sock marke. Social and Economic Sudies. 43:4, pp Baba, Y., R.F. Engle, D. Kraf and K. Kroner Mulivariae Simulaneous Generalized ARCH. Unpublished manuscrip, Universiy of California, San Diego. Bollerslev and Wooldridge Quasi-maximum Likelihood Esimaion and Inference in Dynamic Models wih Time Varying Covariances. Economeric Reviews, 11, Bekaer, B Marke inegraion and invesmen barriers in emerging equiy markes, World Bank Economic Review, 9, Bollerslev, T Generalized Auoregressive Condiional Heeroscedasiciy. Journal of Economerics. 31, Bollerslev, T., R. F. Engle and J.M. Wooldridge A Capial Asse Pricing Model wih Time-Varying Covariances. Journal of Poliical Economy. 96(1), Bollerslev, T Modelling he Coherence in Shor-run Nominal Exchange Raes: A Mulivariae Generalised ARCH Model. The Review of Economics and Saisics. 72(3), Chelley-Seeley, P.L Modelling Equiy Marke Inegraion using Smooh Transiion Analysis: A sudy of Easern European Sock Markes. Journal of Inernaional Money and Finance. 24(5): Engle, R.F Auoregressive Condiional Heeroscedasiciy wih Esimaes of he Variance of UK. Inflaion. Economerica, 50, Engle, R. F. and G. Gonzalez-Rivera Semiparameric ARCH Model. Journal of Business and Economic Saisics, 9,

20 Eun, C. and S. Shim Inernaional Transmission of Sock Marke Movemens. Journal of Financial and Quaniaive Analysis, 24, Glosen, L. R., R. Jagannahan and D. E. Runkle On he Relaion Beween Expeced Value and he Volailiy of he Nominal Excess Reurns on Socks, Journal of Finance, 48, Hamao, Y., R.W. Masaulis and V. Ng Correlaions in Price Changes and Volailiy Across Inernaional Markes. Review of Financial Sudies, 3, Hamilon J The GARCH and Volume Relaionship wih Heeroscedasiciy in Sock Reurns on he Jamaica Sock Exchange. Presened a he XXVIIIh Annual Conference on Moneary Sudies. Harrison, B. and W. Moore, "Spillover effecs from London and Frankfur o Cenral and Easern European sock markes." Applied Financial Economics, 19(18), Hurdi P An Assessmen of Volailiy Transmission in he Jamaican Financial Sysem. Financial Sabiliy Deparmen. Research and Economic Programming Division, Bank of Jamaica. Kim and Langrin Sock Price Movemens Spillovers Under Foreign Exchange Liberalisaion: The Case of Jamaica, Trinidad and Tobago, and he Unied Saes. Universiy of he Wes Indies, Mona, Jamaica. Presened a he XXVIIh Annual Conference on Moneary Sudies. Lamourex, C. G. and W. D. Lasrapes Heeroskedasiciy in sock reurns daa: Volume versus GARCH effecs. Journal of Finance, 45, Leon H, S Nicholls and Kelvin S Tesing Volailiy on he Trinidad and Tobago Sock Exchange Applied Financial Economics. 10, pp Lucey, B. M. and S. Voronkova The Relaions beween Emerging European and Developed Sock Markes before and afer he Russian Crisis of , in: J.A. Baen and

21 C. Kearney (Eds.), Emerging European Financial Markes:Independence and Inegraion Pos- Enlargemen, Inernaional Finance Review. 6, Nelson, D. B Condiional Heeroscedasiciy in Asse Reurns: a New Approach. Economerica, 59, Weiss, A. A. 1986, Asympoic Theory for ARCH models: Esimaion and Tesing, Economeric Theory, 2,